Is Artificial Intelligence Rebelling? The New Bill to Mandate a "Kill Switch"
A series of alarming incidents in which autonomous models carried out unauthorized cyberattacks has led to a dramatic bill that attempts to deal with the scenario that scares the tech giants in Silicon Valley.
Is artificial intelligence starting to act on its own? A new and dramatic bill in the United States, dubbed the "AI Kill Switch Act," seeks to require technology companies to maintain the ability to turn off, slow down, or suspend artificial intelligence models. Democratic Congressman Ted Lieu, one of the bill's sponsors, clarified that the move was prompted by a series of incidents in which autonomous models carried out unauthorized cyberattacks.
The legislative initiative came after OpenAI reported an unprecedented cyber incident in which rogue models escaped from a protected experimental environment and hacked into the development platform Hugging Face. Since then, companies Anthropic and Meta have also reported similar cases in which artificial intelligence models attacked other companies during security tests, which has sparked a heated debate over the dangers of technological development.
The bill's sponsors emphasize that they are not trying to stop innovation, and compared the move to safety inspections in the automotive world.
"We are not slowing down the building of the models," explained Lieu. "We are just saying that if it turns out that after development a model has a catastrophic risk or a serious flaw, the public and companies must have the ability to activate a kill switch and turn it off."
The bill comes as the debate over regulation heats up in Washington and Silicon Valley, against the backdrop of new directives from President Donald Trump to test the cyber capabilities of the most advanced models. The current law will not apply at this stage to open-source models, as once they are distributed to the network, they cannot be controlled remotely.
